Skip to main content

Full text of "Fear of flying?--economic analyses of airline safety"

See other formats






Nancy L. Rose 
MIT Sloan School of Management 


June 1991 







Nancy L. Rose 
MIT Sloan School of Management 

WP#3321-91-EFA June 1991 


Nancy L. Rose 

Associate Professor of Applied Economics 
MIT Sloan School of Management and NBER 

June 1991 

This paper was prepared for a Journal of Economic Perspectives 
symposium on the airline industry. 

MAR 3 1992 



The safety of the conunercial airline industry has attracted considerable 
public attention and debate since economic deregulation of the industry in 1978. 
These concerns have energized economic research on three aspects of airline 
safety. First, has the level of airline safety declined since deregulation? 
Research on this topic investigates whether heightened public concerns about air 
safety derive from objective increases in accident risks. Second, what accounts 
for differences in safety performance across carriers? This literature analyzes 
heterogeneity in carriers' safety records as a means of learning about factors 
that influence safety performance. Third, how do markets respond to airline 
accidents? This work explores the effectiveness of market incentives in 
constraining the safety provision of firms. This paper describes our progress 
in answering each of these queries . 

Professor Nancy L. Rose 
MIT Sloan School of Management 
50 Memorial Drive, Room E52-434 
Cambridge, MA 02139 

The safety of the commercial airline industry has been of 
long-standing interest to policy-makers and the general public. 
This issue attracted particular attention in the wake of airline 
deregulation, amid growing concerns that the historical 
superiority of U.S. jet carriers' safety records may have been 
inextricably linked to economic regulation of the industry by the 
Civil Aeronautics Board. After all, economists argued that the 
suppression of price competition led airlines to focus on service 
competition, and public perceptions of service quality suggest 
substantial reductions in at least some dimensions. Perhaps less 
observable dimensions of product quality, such as safety, have 
experienced equivalent or greater declines. If this were the 
case, traditional measures of welfare gains from deregulation 
could be greatly exaggerated. 

These worries have energized economic research on a broad 
range of issues relating to airline safety. Three questions have 
attracted the most attention from economists. First, has airline 
safety declined since deregulation? Research on this topic 
investigates whether heightened public concerns about air safety 
derive from objective increases in accident risks. Second, what 
accounts for differences in safety performance across carriers? 
This literature analyzes heterogeneity in carriers' safety 
records as a means of learning about factors that influence 
safety performance. It extends the before-and-after deregulation 
research by examining through what links, if any, we might expect 
economic regulation to affect aggregate safety. Third, how do 
markets respond to airline accidents? This work explores the 

effectiveness of market incentives in constraining the safety 
provision of firms. If consumers and insurance companies 
penalize airlines with worse safety records, carriers may be 
disinclined to reduce safety investment, even if regulatory 
changes would permit them to do so. I describe below our 
progress in answering these queries. 

1. Has airline safety declined since deregulation? 

Aggregate statistics on U.S. airline safety provide 
reassurance for travellers concerned that deregulation increased 
the risks of air travel. Virtually all measures of accident or 
fatality risk suggest that the long-term trend toward increased 
airline safety has continued since economic deregulation of the 
airline industry in 1978. This is illustrated in figure 1, which 
plots the number of aircraft accidents per million departures for 
large U.S. scheduled air carriers over the period 1955-1990.^ 
Both total and fatal accidents per million departures declined 
substantially, although there is considerable variation in 
accident rates from year to year. 

There is little evidence that improvements in airline safety 
have slowed appreciably since deregulation. Observed accident 
rates since 1978 conform closely to those predicted by a trend 
estimated over the 1955-1977 data, as illustrated in figure 2. 
More formally, regression analysis of the log of accident rates 

^ Referred to as "Part 121" carriers, these are carriers that 
operate aircraft with capacity in excess of 60 seats. These 
carriers currently operate primarily jet aircraft fleets. 

on a time trend indicates that the coefficients on either a 
deregulation dummy variable or a variable measuring time since 
deregulation are insignificantly different from zero.^ Figure 2 
does, however, suggest some scope for caution. Accident rates 
over the last four years (1987-1990) lie slightly above trend. 
There is not enough data to determine whether this reflects 
normal variation in observed accident rates over short time 
horizons or an elevation of the true underlying risk, nor is it 
obvious that effects that do not materialize until ten years 
after deregulation should be attributed to regulatory changes 
rather than to some other cause. Nevertheless, these data may 
suggest continued scrutiny of aggregate safety performance over 
the next few years. 

Passenger fatality rates also exhibit continued improvement 
after deregulation. For example, Barnett and Higgins (1989) 
calculate that fatality risks for passengers on U.S. domestic jet 
airline flights declined from an average of 1 in 2.5 million 
flights over 1971-78 to 1 in 7.4 million flights over 1979-86. 
They argue, however, that the decline in risk would have been 
even greater, but for the entry of new jet carriers post-1978. 
As evidence, they separate the U.S. carriers into "established 
carriers" (trunk and local service airlines existing as of 1978) 
and new entrants (a group of 19 "jet children" of deregulation, 

2 Rose (1989) presents results for 1955-1986 data; my updates 
based on the 1955-1990 data yield similarly insignificant results. 
This conclusion is reinforced by an analysis of the time fixed 
effects (1958-1986) from the model of airline specific accident 
counts in Rose (1990). 

most now out of business) . For 1979-86, fatality risk for 
passengers on established carriers averaged 1 per 11.8 million 
flights. In contrast, the group of entrants Barnett and Higgins 
analyze had an aggregate fatality risk of 1 per 870,000 flights! 
This does not imply that the planes of the entrant carriers were 
continually dropping out of the sky, however: only 3 of the 19 
carriers had any domestic passenger fatalities during the 7 year 
period, and these had just one fatal accident each. The high 
risk arises from the fact that the entrants carried relatively 
few passengers. The robustness of this conclusion and the safety 
records of entrants will be discussed further when we analyze 
differences in safety performance across carriers. 

Analyses of the causes of airline accident rates can shed 
additional light on the effects of deregulation. If deregulation 
induced carriers to cut maintenance activities, for example, one 
might expect to observe more accidents due to equipment failure. 
Accidents due to pilot error should increase if airlines 
compromised safety by hiring less experienced pilots, reducing 
training, or working pilots harder. If increased congestion, 
combined with the reductions in air traffic control (ATC) staff 
after the 1981 controllers strike, degraded the air traffic 
control system, accidents resulting from ATC errors or 
interference by other aircraft should become more common. 

To test whether deregulation has had these effects, Oster 
and Zorn (1989) analyze National Transportation Safety Board 
(NTSB) Accident Briefs for scheduled domestic passenger service 

accidents over the 1971 through 1985 period. For each accident, 
they select as the "primary cause" the event or action that 
initiated the sequence of events culminating in the accident.-^ 
These causes are then grouped into categories that might be 
sensitive to deregulation-induced changes, such as Pilot Error, 
Equipment Failure, Air Traffic Control Error, and Other Aircraft 
(General Aviation) , and categories that are unlikely to be 
influenced by deregulation, including Weather, Seatbelt Not 
Fastened, and Other. Between the regulated (1970-78) and 
deregulated (1979-85) periods, total accidents per million 
departures for trunk and local service carriers declined by 54%. 
Accident rates due to equipment failure, pilot error, ATC error, 
and other aircraft declined by this amount or more, topped by a 
71% reduction in accidents initiated by equipment failures. This 
suggests a relative decrease in accidents due to causes under a 
carrier's control after deregulation. 

Further evidence on the changes in maintenance practices and 
their effects on safety since deregulation is provided by 
Rennet's (1990) study of jet engine maintenance histories. 
Kennet analyzes complete aircraft engine histories for 42 Pratt 
and Whitney jet engines, operated by 7 different airlines. He 
finds that the length of time between maintenance shop visits has 
increased since deregulation, but that deregulation has had no 

' Because their criterion differs from that used by the NTSB, 
their distribution of accidents by cause differs from the NTSB 
distribution. Broadly similar conclusions are reached by Morrison 
and Winston (1988) , who analyze the distribution of fatal accidents 
using NTSB causes. 

effect on the probability of an engine shutdown. This may 
reflect a drive toward more efficient maintenance policies and 
practices in the wake of deregulation. The result that engine 
shutdown probabilities have been unaffected suggests that these 
maintenance changes have not compromised air safety, consistent 
with Oster and Zorn's report of substantial relative declines in 
accidents initiated by equipment failure. 
Indirect effects of deregulation on travellers^ safety 

There are a number of indirect channels through which 
deregulation may have influenced safety. First, the shift from 
jet airline to commuter airline service in many small communities 
may have increased risks for passengers on these routes. Second, 
increased reliance on hub-and-spoke networks may have increased 
the average number of stops or plane changes passengers must 
make. Since accident risks are roughly proportional to the 
number of take-offs, this would tend to increase passengers' 
risks per trip (origin-destination) . Third, the introduction of 
price competition and service improvements may induce travellers 
to substitute air travel for auto travel. Since the risk of a 
highway accident substantially exceeds that for air travel over 
even moderate distances, this substitution would enhance 
travellers' safety. 

Substitution of commuter service : By eliminating explicit 
cross-subsidization and easing entry and exit restrictions, 
airline deregulation may have encouraged established jet carriers 
to abandon uneconomic service to small communities. While most 

of these communities retain air service, it now typically is 
provided by commuter carriers. Because commuter airlines have 
higher accident rates than jet airlines, risks to travellers in 
these communities may have increased. For example, over 1979- 

1985, passenger fatalities were .38 per million passengers 
enplaned on trunk airlines, but 1.27 per million passengers 
enplaned on commuter airlines — more than three times greater for 
commuters (Oster and Zorn, 1989) . 

These simple comparisons may substantially overstate the 
change in risk, however (Oster and Zorn, 1989) . First, the 
largest commuter airlines are much safer than the smaller 
commuters, and these are the ones that typically have replaced 
jet carriers. The top 2 commuters, for instance, had passenger 
fatalities of .67 per million enplanements, roughly half the risk 
for commuters overall.^ Second, service substantially improved 
on the routes where commuters replaced jets, with fewer inter- 
mediate stops and more weekly departures. In a sample of 60 
city-pair markets where commuters replaced jets between 1978 and 

1986, the average number of intermediate stops fell by half (from 
.59 to .30; see Oster and Zorn, 1989). Re-scaling the fatality 
risk to reflect total risk per passenger trip on these routes 
yields a risk of .60 per million trips for jet carriers (.38 

* There have been no studies that look at commuter safety 
under codesharing arrangements with major carriers (see Borenstein, 
1991, for a discussion of codesharing) . Given the increased 
scrutiny that codesharing imposes upon the commuters, it is likely 
that their safety record is even better than implied by size alone. 

fatalities per million enplanements times 1.59 average take-offs) 
compared to a risk of .87 for the large commuters (.67 fatalities 
per million enplanements times 1.30 average take-offs). While 
the commuter risks are higher, the differences are less stark 
than implied by the initial comparison. 

Finally, the average weekly departures in these 60 markets 
more than doubled after commuters took over service (from 2.88 to 
6.29). The increased frequency of service appears to be 
associated with increased ridership, at least part of which 
reflects a switch from cars to planes for some travellers. Oster 
and Zorn (1989) estimate the auto fatality rates in these markets 
to lie between 1.9 and 2.3 per million passenger trips. Since 
this is substantially greater than the risk for the larger 
commuter airlines, the modal switch enhances overall safety for 
these travellers. 

Increases in the average number of stops per trip ; The 
second potential indirect effect of deregulation, possible 
increases in the number of stops or plane changes passengers must 
make en route to their final destination, has not been well- 
documented. While the development of hub-and-spoke networks may 
substitute one-stop or one-change service for nonstop service in 
outlying markets, it is likely to increase nonstop service 
availability for passengers travelling to and from the hub. The 
net impact on average stops cannot be predicted a priori . 

Some evidence on this effect is provided by Borenstein 
(1991). He finds an increase in the number of passenger trips 


that involve a change of plane, from 27.3% of trips in 1978 to 
32.8% in 1990. If all remaining passengers flew nonstop, the 
average number of flights per trip would have increased by 4.3% 
over this period (from 1.273 to 1.328). While this increases air 
travel risks, the overall impact is not substantial. The average 
total (fatal and nonfatal) accident rate per million flights 
declined by 54% between the 10 years prior to deregulation and 
the 10 years after deregulation. Adjusting for a 4% increase in 
average flights per trip reduces the effective decline to 52%. 

In fact, direct (no change of plane) service includes both 
nonstop and one- (or multi-) stop flights. Because there have 
been no studies of the change in the average number of stops for 
these passengers, we cannot determine the overall change in 
average departures per trip. Based on the results for the change 
of plane statistics, however, failure to account for this seems 
unlikely to alter the basic conclusion. 

Shifting traffic from highways to air ; The lower average 
fares and the widespread adoption of discount fares and 
sophisticated price discrimination schemes that resulted from 
deregulation substantially increased air travel. Between 1975 
and 1985, domestic passenger enplanements for the largest U.S. 
carriers grew at a rate of 6.6% per year and domestic revenue 
passenger-miles (RPMs) grew at 7.5% per year. Some of this 
increase represents new travel, that is, trips that otherwise 
would not have taken place. Some of the increase represents a 
shift from other modes of travel, such as automobile, rail, or 


It is difficult to determine the precise extent to which 
travellers have shifted from automobile travel to air travel as a 
result of airline deregulation. Using annual aggregate data on 
passenger car miles travelled and a dummy variable for airline 
deregulation, McKenzie and Warner (1988) estimate a decline of 
nearly 4% in passenger car miles as a result of airline 
deregulation, or an average reduction of 43 billion car miles 
annually during the 1979-85 period. They conclude that this 
reduction in auto miles corresponds to roughly 1700 fewer auto 
fatalities per year. If the average auto occupancy rate for 
intercity traffic is 2.0, a shift of 43 billion car miles to air 
travel would imply an increase of 86 billion passenger miles for 
airlines. The number of air fatalities associated with this 
amount of air travel averages about 41. A shift of this 
magnitude from highway to air would have an enormous net savings 
in lives: more than 1650 per year. Is this a credible estimate? 

Airline RPMs increased by roughly 70 billion between 1975 
and 1980, or 140 billion between 1975 and 1985. If the estimated 
shift in highway travel is correct, the bulk of the increase in 
air RPMs comes from displaced auto trips. This seems implausibly 
large. Unfortunately, we do not have better estimates of the 
true magnitude of the modal shift. Even if the effect is only 
one-fifth as large as McKenzie and Warner estimate, however, more 
than 300 lives would be saved each year by the shift to air 


travel — more than the total U.S. airline passenger fatalities in 
any of the last 10 years. 

2. What accounts for differences in accident rates across 

Against the backdrop of substantial declines in aggregate 
accident rates over time lie wide variations in accident rates 
across individual carriers within any time period. Figures 3 and 
4 illustrate this in histograms of total accident rates per 
million departures for a sample of major airlines over the 1971- 
75 and 1981-85 periods, respectively.^ The wide variation in 
individual accident rates is not entirely surprising: given the 
discrete and infrequent nature of accidents, one additional 
accident in a five year period can generate an enormous increase 
in a typical airline's accident rate per million departures. 
This raises the question: do these statistics reflect expected 
random fluctuations around a common mean accident rate or more 
systematic differences in behavior and subsequent safety 
performance across airlines? 

Economists have concentrated their efforts to model 
differences in carriers' safety records in three areas: the 
impact of airlines' financial condition on their safety 
performance, variations in safety performance between entrants 

^ These plots are based on data for a sample of 35 large 
airlines, as reported in Rose (1990) . The 1981-85 plot omits World 
Airlines, which had two accidents and an accident rate of more than 
51 per million departures. The next highest accident rate was 12.5 
per million departures. 


and established carriers, and the determinants of higher accident 
rates for commuter carriers relative to jet airlines. 
Financial impacts on airline safety 

The potential impact of financial pressures on airlines' 
safety performance has provoked a long-standing debate in policy 
circles and attracted particular attention since deregulation. 
The argument that competition has reduced profit margins and 
forced carriers to "cut corners" on safety has been one of the 
key weapons in the arsenal of re-regulation advocates. A variety 
of economic models can generate predictions consistent with a 
financial link to safety, including models of reputation 
formation under asymmetric information, liquidity constraints on 
investment behavior, and firm decision-making near bankruptcy. 
None of these models implies that such a link must exist, 
however, leaving the resolution of this debate to empirical 

Early studies, typically based on short time series for 
small cross sections of carriers (or industry aggregate time 
series regressions) , detected no significant relationship between 
financial variables such as profitability and airline accident 
rates. For example, Golbe (1986), who looked at cross-sections 
of 11 domestic trunks over the 1963-66 and 1967-70 periods, found 
an insignificant positive relation between profitability and 
accident rates. These studies share a common shortcoming, 
however: the infrequent nature of airline accidents combined 
with their small sample sizes may limit the power of their 


statistical tests. 

Analyses of more extensive data sets and alternative safety 
measures find evidence that lower profit margins are associated 
with worse safety performance, at least for some groups of 
carriers. Rose (1990) explored the determinants of airline 
safety performance for a panel of 35 part 121 U.S. carriers over 
1957-1986. In the full sample, higher operating profits were 
associated with lower accident rates in the following year. A 5 
percentage point increase in the operating margin (e.g. from 5% 
to 10%) implies about a 5% reduction in the total accident rate 
and more than a 15% reduction in the fatal accident rate, other 
things equal. This result for total accidents is replicated by 
Evans (1989) in a study of accident rates for nearly 100 carriers 
over 1970-87. 

These average effects may themselves mask important 
differences across carriers in the sensitivity of safety 
performance to profitability changes. Rose's data suggest that 
profitability effects may be strongest for the smaller and mid- 
size carriers in the sample, and may not be important for the 
very largest carriers studied. This pattern is particularly 
clear in the analysis of airline incidents, in which higher 
profits are associated with lower reported incidents for small 
and mid-size carriers, but higher incident rates for the very 
largest carriers. A 5 percentage point increase in the operating 
margin implies about a 20% reduction in reported incidents for 
the smallest carriers in the sample and a 10% - 12% reduction for 


mid-size carriers. 

The strength of the profitability-safety link for the small 
and mid-size carriers may indicate greater flexibility in these 
firms' safety investment choices. A number of factors could make 
the safety investment levels of large firms less variable: 
public information about underlying safety levels may be better 
for the largest airlines (reducing information asymmetries) , 
large airlines may have better access to capital markets or 
"deeper pockets" for internal financing, and FAA regulators may 
more closely scrutinize these carriers. This heterogeneity also 
may help to explain why the earlier studies, which tended to 
focus only on the very largest (trunk) carriers, failed to detect 
a link between profitability and safety performance. 

A significant remaining gap in our analysis of financial 
influences on safety is an understanding of the profitability 
effects for the very smallest air carriers in the industry; 
commuter carriers. While recent studies include a much broader 
range of carriers than had previously been studied, they continue 
to be limited to "jet" (Part 121) carriers due to the lack of 
reliable financial data for commuter (Part 135) carriers. 
Anecdotal evidence suggests that commuters may be quite sensitive 
to financial pressures, and the argioments raised above for the 
smaller jet carriers would seem to apply even more strongly to 
commuters. Decisive conclusions about this segment of the 
industry must await further data and study, however. 
New entrant safety performance and the role of experience 


A major concern after deregulation was the safety 
performance of new entrants into the airline industry. Barnett 
and Higgins' (1989) conclusion that entrant carriers were 
substantially more risky than established carriers in terms of 
passenger fatalities heightens this concern. The empirical 
evidence on this issue is somewhat mixed, however. The relative 
riskiness of entrants appears sensitive to the measures of safety 
performance employed in the study, and also may depend on the 
definition of entrant carriers and identities of the firms 
included in the sample. 

The most thoroughly studied measure of safety performance 
for new entrants is total accidents per million aircraft 
departures. Virtually all analyses using this measure of safety 
indicate that entrants do not perform significantly worse than 
established carriers (e.g., Kanafani and Keeler, 1989; Oster and 
Zorn, 1989; and Evans, 1989). Kanafani and Keeler (1989), for 
example, find that identifying a carrier as an entrant does not 
add significant explanatory power to a regression model of total 
accident rates over 1982-85, perhaps in part because of the 
enormous variability in accident rates across the 25 entrants in 
their sample. Evans (1989) argues that entrants appear to have 
lower accident rates than established carriers, other things 
equal. His analysis of 105 carriers over 1971-1987 suggests that 
post-deregulation entrants have accident rates that are roughly 


half those of established carriers, other things equal. ^ This 
result is not sensitive to whether the entrants are defined as 
completely new airlines or include carriers that previously 
provided intrastate or charter service. Evans argues that this 
result may reflect more intense regulatory scrutiny of airlines 
newly certified in interstate service. 

The general conclusion that entrant safety performance does 
not significantly differ from that of established carriers holds 
across a wide variety of safety measures. Oster and Zorn (1989) 
find no significant differences between trunks and "other jet 
carriers" for five of six aggregate safety measures over 1979-85, 
including passenger fatalities and passenger injuries per million 
enplanements, and total accidents, serious injury accidents, and 
minor accidents per million aircraft departures. Their group of 
"others" corresponds to the broadest definition of entrants used 
in the literature. Kanafani and Keeler (1989) report no 
significant difference in FAA inspection ratings for new entrants 
under the National Air Transportation Inspection program and some 
evidence that new entrants have lower near mid air collision 
reporting rates than do established carriers (though the latter 
may reflect differences in reporting incidence rather than 

^ The relative accident rate for entrants in Evans's study 
should be calculated as exp(NEW - DEREG) , where NEW is a dummy 
variable for new entrants (estimated at about -1.3) and DEREG is a 
dummy variable for established carriers post-1978 (estimated at 
about -.50). This calculation yields the value .44, implying that 
entrant accident rates are 44% of established carrier accident 
rates, other things equal. Note that this is not the calculation 
apparently reported by Evans. 


differences in occurrence rates) . 

The dominant exceptions to this sanguine view of new 
entrants are based on analyses of fatal accident rates. In 
addition to the Barnett and Higgins (1989) analysis discussed 
earlier, Oster and Zorn (1989) report that entrants (their "other 
jet carriers") had a substantially higher aggregate rate of fatal 
accidents per million departures over 1979-85 (.90 v. .22 for 
trunk and local service carriers) . As noted earlier, this poor 
aggregate performance masks substantial heterogeneity across 
carriers, with most entrants massed at zero fatalities and a few 
extreme outliers pulling up the aggregate fatality rate. 

Unfortunately, there have been no carrier specific analyses 
of fatal accident rates to discern the sensitivity of the 
conclusions to this heterogeneity or to the definition of entrant 
airlines. For example. World Airlines, which had two accidents 
and a fatal accident rate of 51 per million departures over 1981- 
85, is included as an entrant in studies of entrant fatality 
risk. While the airline was new to scheduled interstate service, 
it had been operating charter service prior to deregulation. 
Should World be grouped either with People Express, which entered 
airline service de novo after deregulation, or with Pacific 
Southwest Airlines, which had provided California intrastate 
service since 1948? In most studies, "entrants" are defined to 
include all of these types of carriers. 

To understand which firms can be meaningfully grouped 
together, we must first understand the possible underlying causes 


of the entrant results. This is difficult to do with either 
aggregate analyses or simple dummy variable regressions of 
carrier differences. Unfortunately, few studies have attempted 
to move beyond these approaches. Oster and Zorn (1989) report 
that entrants as a group have a higher total accident rate 
attributable to pilot error (.60 per million departures, compared 
to .16 for trunks). This might be consistent with entrants' 
pilots being on average less experienced or less well-trained, 
either overall or relative to their new positions. Rose (1990) 
provides evidence of some general learning-by-doing effects on 
safety performance. For total accident rates, airline operating 
experience has at most a weak negative effect, which vanishes in 
specifications that control for a carrier's average accident 
rate. For both fatal accidents and total incidents, however, 
experience exerts a strong, statistically significant negative 
effect: more experienced airlines have fewer fatal accidents and 
fewer incidents, other things equal. Although these estimates 
are not based solely on entrant performance, the results are 
broadly consistent with studies that find no significant entrant 
effect for total accident rates, but worse entrant performance on 
fatal accidents. Additional investigation is required to 
develop a better understanding of other sources of the apparent 
differences in safety performance between entrants and 
established carriers. 
Commuter carriers 

Commuter airlines, as a group, have substantially higher 


accident and fatality rates than do jet carriers. The 
implications of this observation depend critically upon the 
source of these differences. For example, if commuter airlines 
invest less in safety, other things equal, then more rigorous FAA 
regulation of their safety practices would tend to improve their 
safety records.^ Such regulation will have little effect if the 
disparities arise from inherent differences in equipment 
reliability (e.g., smaller, propeller aircraft are more prone to 
failure, even when optimally equipped and maintained) or airport 
facilities (e.g., commuters are more likely to serve airports 
that lack advanced navigational aids or offer more hazardous 
operating conditions) . Similarly, if most of the performance 
differences are attributable to route rather than carrier 
conditions, then sxibstituting one type of carrier for another on 
a given route is unlikely to have much impact on safety. 

Discerning the relative importance of carrier and route 
conditions on commuter safety records would be difficult under 
any circumstances. This task is further impeded by the dearth of 
reliable, detailed firm level data for this segment of the 
industry. Nevertheless, there is suggestive evidence that 
carrier investment has a substantial impact on safety performance 
in this sector. First, commuters that were part of the Allegheny 
(USAir) commuter system had an overall safety record that matched 
the jet carrier safety record over the 1970-80 period, despite 

^ Whether this is socially optimal depends on whether 
commuters currently underprovide safety. 


substantially higher accident rates for the commuter industry as 
a whole (Meyer and Oster, 1987) . This is unlikely to be solely 
attributable to differences in the routes and equipment of these 

Second, in 1978 the FAA substantially tightened commuter 
safety regulations, increasing pilot qualification, crew 
training, and maintenance requirements (particularly for larger 
commuter aircraft) , and specifying for the first time minimum 
equipment lists for commuter flights. This appears to have had a 
dramatic impact on aggregate commuter safety. The commuter 
passenger fatality rate per million enplanements declined by more 
than half between 1970-78 and 1979-85, with the bulk of the 
decline occurring in accidents caused by equipment failure, pilot 
error, and weather (the latter presumably influenced by both 
enhanced pilot certification and training requirements and 
equipment rules governing instrument flight rule operations; see 
Oster and Zorn, 1989) . Since commuter regulations remain less 
stringent than those for jet carriers, additional improvements in 
safety are likely to be possible — although whether these would 
be welfare enhancing remains unknown. 

3 . How do markets respond to airline accidents? 

For air travellers, safety is an important aspect of product 
quality. Unlike other characteristics of product quality, such 
as schedule convenience, crowding, and on-board service, 
consumers have difficulty observing air carrier safety levels 


when they make their travel decisions. As in other markets where 
consumers cannot observe or evaluate product characteristics, 
there is reason to suspect that the market may supply less safety 
than consumers would demand if fully informed. Concern with 
potential market failure has led to a complex web of government 
regulations that specify minimum safety input and performance 
standards for air carriers. Airlines' and aircraft 
manufacturers' reputations may provide an alternative (or 
complementary) mechanism for insuring adequate safety provision. 
If these are effective checks on behavior, we should observe 
market penalties for firms that deviate from their established 
reputations. This notion has given rise to a substantial 
economics literature that evaluates market responses to air 
carrier accidents. 

We can analyze market responses to airline accidents from 
two perspectives. First, does the market penalize aircraft types 
involved in an accident: what is the effect of an accident on 
the profits of the aircraft's manufacturer, the profits of 
airlines that operate a substantial number of that aircraft type, 
and the traffic patterns of passengers who previously flew on 
that aircraft type? Questions of this sort will be most 
appropriate when flaws in the aircraft itself are suspected to 
have contributed to a particular accident. Second, does the 
market penalize airlines that are involved in accidents: how 
does an accident affect an airline's profits and traffic flows, 
and the profits and traffic flows of its competitors? These 


questions will be most appropriate when an airline's actions or 
inaction are suspected to have contributed to the accident. 

In this literature, profit effects typically are measured 
using an event study methodology, which measures the change in 
the equity share price of a firm following an accident. This 
yields an estimate of the expected change in the present 
discounted value of future profits resulting from the accident. 
Traffic responses have been analyzed both by examining changes in 
"before and after" market shares and by measuring the deviation 
from predicted demand using econometric models of airline demand 
functions. The samples are restricted to fatal accidents, and 
most studies exclude cargo and crew only (re-positioning) 
flights. These criteria select the worse and more highly 
publicized accidents for analysis. 
Aircraft reputation 

Studies of aircraft reputation effects have focused on two 
DC-10 crashes: the American Airlines Chicago crash on May 25, 
1979, which is the worst domestic U.S. airline accident (273 
fatalities), and the United Airlines Sioux City crash on July 19, 
1989 (Barnett and LoFaso, 1983; Chalk, 1986; Karels, 1989; 
Barnett, Menighetti, and Prete, 1990) . Both of these accidents 
raised concerns about potential DC-10 manufacturing or design 
problems. One study (Chalk, 1987) also examines accident effects 
on aircraft manufacturers' profits across a broader sample of 


"suspect" crashes. ° What do these analyses reveal? 

The 1979 DC-10 crash provides some evidence of a market 
penalty for aircraft manufacturers. McDonnell Douglas, the 
manufacturer of the DC-10, lost roughly 10 percent of its equity 
market value, or approximately $100 million, in the first four 
days after the accident.' The firm's shares declined by an 
additional 10 percent when the FAA announced its unprecedented 
decertification of the DC-10, an action that grounded the entire 
DC-10 fleet indefinitely. These market value declines are 
substantially larger than any direct costs imposed by the 
accident, and would be consistent with lower expected sales of 
McDonnell Douglas aircraft as a result of the accident. ^° 

These declines are not representative of responses to other 
accidents, however. In contrast to the 1979 experience, 
McDonnell Douglas appears to have been unaffected by the 1989 
Sioux City accident. Despite early reports that the design of 

^ Chance and Ferris (1987) find no effect on the manufacturer 
for a sample of 46 accidents over the 1962-1985 period. Their 
sample is not, however, stratified by likely cause of the accident. 

' The accident occurred after the market close on Friday, May 
25, of Memorial Day weekend. The share price response therefore is 
measured from the Friday close to the Tuesday close. See Chalk 
(1986) . 

^° As new information suggested that improper maintenance 
practices were the likely cause of the accident, at least part of 
the initial share price declines were reversed. The exact 
estimates of the net effect on McDonnell Douglas appear highly 
dependent on the time period over which stock returns are 
evaluated. Chalk (1986) reports statistically significant net 
declines of 14 to 22 percent through various dates in July 1979. 
Karels' (1989) attempts to reproduce these results yielded 
estimates of +1 through -21 percent net returns, all statistically 
insignificant . 


the DC-10 hydraulic system was a major factor in the crash, 
returns on McDonnell Douglas stock were commensurate with market 
returns over the days following the accident." Chalk's (1987) 
evidence on manufacturer losses for a sample of 19 accidents to 
which aircraft failures contributed suggests modest profit 
losses, but these estimates may be strongly affected by the 
inclusion of the 1979 DC-10 crash in the sample. Chalk finds an 
average share price decline of roughly 4% over the five business 
days following an accident, corresponding to an average loss of 
$21 million in market value. His data indicate no statistically 
significant share price effects for accidents involving Boeing or 
Lockheed aircraft, however, and the estimated average McDonnell 
Douglas decline is likely to be quite skewed by the massive 
declines associated with the 1979 crash. 

Profit declines for aircraft manufacturers do not appear to 
result from passenger avoidance of aircraft involved in fatal 
accidents. Barnett and Lofaso's (1983) study of DC-10 market 
shares 6 months before and 6 months after the 1979 crash revealed 
no systematic changes in travellers' behavior on a sample of 18 
routes. ^^ In a study of travel agency ticketing data, Barnett, 

" The accident occurred after the market closed on July 19; 
July 20 therefore is the first post-crash return day. McDonnell 
Douglas shares lost nearly 7% on July 19, probably due to its 
announcement on that day of unexpectedly large second quarter 
losses. On July 20, McDonnell Douglas shares declined 0.9%, 
compared to a 0.7% for the market as a whole. McDonnell Douglas 
share prices rose over the next week. 

^^ While Barnett and Lofaso control for some airline route 
characteristics, they do not have data on average fares. It is 
possible that airlines with DC-10 service lowered fares to retain 


Menighetti, and Prete (1990) find evidence of very short-term DC- 
10 avoidance following the 1989 Sioux City crash. In their 
sample of 14 routes, 1 in 3 passengers who booked travel within 
the first 2 weeks after the crash avoided choosing DC-10 flights, 
relative to pre-crash behavior. This behavior quickly 
dissipated, however, with booking shares returning to within 10% 
of pre-crash levels by 8 weeks after the crash. ^^ Moreover, 
despite the development of sophisticated pricing and inventory 
management systems by 1989, airlines did not appear to lower 
prices on DC-10 flights in response to initial traffic declines. 

Finally, there is some evidence that the 1979 DC-10 crash 
adversely affected airlines that owned substantial numbers of 
these aircraft, although there have been no general studies of 
this effect. Karels (1989) finds share price declines for both 
American Airlines (the operator involved in the crash) and a 
portfolio of other airlines operating DC-lOs in the aftermath of 
the 1979 accident. The first response to the crash was a 2% 
decline in share values, although this could not be statistically 
distinguished from zero. The decertification announcement led to 
a 5.3% decline for American and a 2.9% decline for the other DC- 
10 airlines. A portfolio of non-DC-10 airlines was unaffected. 

How should we interpret these studies? It seems premature 

market shares. The study of the 1989 DC-10 crash suggests that 
this explanation is unlikely to account for their result. 

13 The study did not examine booking patterns beyond 8 weeks 


to cite these as confirmation of a "reputation effect," at least 
in the sense that "market forces can compel producers to invest 
in safety, even if consumers are ignorant of all the technical 
details of the product" (Chalk, 1986) J^ The strongest evidence 
of market responses is associated with the 1979 American Airlines 
DC-10 crash; evidence of market responses to other accidents is 
weak to non-existent. In 1979, however, the market may have been 
responding more to specific FAA interventions than to general 
reputation effects. FAA airworthiness directives can require 
airlines and manufacturers to invest substantial amounts in 
inspections and repairs, replacements, or re-designs of aircraft 
components. The FAA's 1979 decision to revoke the DC-10 's 
certificate grounded the existing fleet of DC-lOs indefinitely 
(inducing direct losses for DC-10 operators) and raised the 
possibility that McDonnell Douglas would be required to make 
extensive modifications as a prerequisite to selling any 
additional aircraft (and re-certifying the existing fleet) . 
While market reputation effects and direct FAA interventions both 
may induce manufacturers to invest in aircraft safety, the policy 
implications of these two mechanisms are quite different. The 
existing empirical evidence does not decisively indicate which 
mechanism is more important. 
Airline reputation 

^^ One should remember that while air passengers may not be 
well informed about technical characteristics, they are only 
indirect consumers of aircraft services. The direct customers of 
aircraft manufacturers are airlines, which tend to be highly 
knowledgeable and sophisticated buyers. 


A number of studies have investigated market responses to 
accidents at the airline level: does an accident reduce the 
airline's expected profitability? Two of the more interesting 
and careful of these analyses are Borenstein and Zimmerman's 
(1988) study, which couples an investigation of profit effects 
with traffic responses, and Mitchell and Maloney's (1989) study, 
which pairs an examination of profit effects for different 
classes of accidents with a study of insurance premia changes. 
Both find evidence of modest profitability declines in response 
to fatal accidents. 

Borenstein and Zimmerman analyze responses to 74 fatal 
accidents over 1962-85. For the 62 accidents that occurred while 
passengers were on board the aircraft, they find an average 
decline in equity value of roughly 1.3% on the first trading day 
following the accident, and 1.5% over the first two days 
following the accident. This translates into an average $12 
million loss in 1990 dollars.''^ Mitchell and Maloney divide 
their sample of 56 accidents over 1964-87 into 34 "pilot error" 
crashes and 22 "carrier not at fault" crashes. For the pilot 
error sample, they find a one day decline of roughly 1.6% and a 
two-day decline of roughly 2.3%.^* This corresponds to an 

^^ All dollar values reported in this section have been 
escalated to 1990 dollars using the CPI. 

^^ The point estimate declines for the carrier not at fault 
sample are about half as large and are quite imprecisely estimated. 
This may suggest, as Mitchell and Maloney conclude, that the market 
does not penalize airlines for accidents not caused by pilot error. 
From a different perspective, however, a pooling test across the 
two samples would not reject the hypothesis that both sets of 


average loss in equity value of $22 to $31 million in 1990 
dollars. Because airlines typically carry quite complete 
hull (aircraft) and liability insurance, most of the equity 
decline appears to arise from prospective losses, rather than 
actual cash outlays resulting from the current accident. Two 
possible sources of prospective losses are increased insurance 
premia and reduced demand due to reputation effects. Mitchell 
and Maloney estimate that the additional liability insurance cost 
over a five year period following an at-fault accident is roughly 
90 percent of the one year premium pre-crash. The total present 
discounted value of insurance increases average about $10 million 
in 1990 dollars. ^^ This accounts for one-third to one-half of 
their estimated decline in equity value. 

Borenstein and Zimmerman investigate the impact of accidents 
on demand for an airline's services. They find virtually no 
effect of an accident on demand during the regulated period of 
their sample (1960-77) . After deregulation, there may be a 
short-term demand response to an accident. In their sample of 13 
accidents over 1978-85, estimates of the total loss in demand 
over a four-month period average 10% to 15% of one-month's 
traffic voliome, although these estimates are at best of marginal 
statistical significance. Consistent with the implications of 

results are drawn from the same distribution. 

^^ Their results for hull insurance increases are quite 
sensitive to the specification of the model. An estimate of hull 
insurance increases is included in the total dollar value of 
insurance increases, however. 


the DC-IO traffic response studies, this decline is <^ite short- 
lived: „ost Of the effect appears to be experienced in the first 
two months following a crash. 

It is difficult to interpret these results. The demand 
Changes during the deregulation era, while relatively small and 
Short-ten., imply large revenue losses. For the sample of 13 
accidents, the average implied revenue loss is over Sloo million 
in 1990 dollars.- This suggests considerable penalties 
for airlines involved in fatal accidents. The strength of this 
conclusion is, however, limited by a number of factors. First 
these results are based on a relatively small sample and are 
estimated very imprecisely, second, the estimated revenue losses 
substantially exceed the estimated declines in eguity value, and 
the difference is unliKely to be accounted for by cost reductions 
associated with serving fewer passengers in the very short-term. 
Thxrd, revenue losses appear to be uncorrelated with the change 
in equity value in this sample. Finally, there is relatively 
little evidence that accidents have a significant effect on the 
demand or profits of an airline's competitors. Over the entire 
deregulation period, Borenstein and Zimmerman's point estimate of 
the demand change for other airlines following an accident is 
negative, but very small and imprecisely estimated. The 8 
largest accidents (lOO or more fatalities) may have induced a 
small (1%, one-month increase in demand for other airlines, but 
the stock price of these airl ines was unaffected. This suggests 

" The uncertainty around this estimate is, however, enormous. 


that most passengers who would have flown an airl ,• 

i-xown an airline recently 
involved in an accident instead choose not to ,lv whi.H 

be entirely plausible Furth ■ "^ "°' 

ible. Further investigation, using the 

additional years o, post-deregulation data now available, appears 
necessary to address these concerns and resolve the ^es J: ^ 
demand effects. 

While the literature suggests th^ .. -u-, • 
penalties . Possibility of some market 

penalties for airlin^^ i-h,^*- 

^•'^'^ experience passenger fatalities 
these .nethodologies may be inherently incapable „, • ' 
definitive tests of the . ^"capable of providing 

Of the strength of aircraft or airline 
reputation ef feoi-c j» • i • 

effects. Airline accidents, while newsworthy „av 

no be very informative, .he expected .or optimal, levelo" 

tnis, the occurrence of =„ 

xrence of an accident may not ca„c=^ ^ 
revic^o ^.u • ^^^ consumers to 

revise their safety expectations for a firm if 
not l^.r, ^"^ accident does 

ea consumers to revise their priors about an aircraft-s or 
- ne. safety, consumers should not penalise the firm involved 
the accident. „ini„., _,„,, ^^^^_^^ ^^ ^^ ^^^^^^^^ 

r " "■' """ "-"^ *"" "^^^^"^ ^- -Pected safety 
-vels would be severely punished and therefore are deterred from 
ever ..cheating., and ineffective reputation mechanisms <e.g 
Where consumers are unaware of the aircraft type used on 
particular flights, have difficulty assessing safety records and 
a Slow to update their priors in response to accidents, or 
Slow to respond to differences in perceived accident ris.s 


acres, aircraft or airlines, . The existing analyses do not 
enable us to distinguish these two extremes. 


Economists have learned a substantial amount about airline 
safety, even though many questions remain unanswered, m fact 
one might wonder about the motivation for devoting so much energy 
to studying such a low risK activity. Mrline safety analyses 
appear to have garnered a disproportionate share of major Journal 
pages in recent years, relative to more economically significant 
riste. m,ile our professional fascination may be inspired in 
part by the amount of time we spend in the air, we are not alone 
m this interest. Airline accidents attract far more public 
attention than most other sources of fatality risK, including 
such popular concerns as cancer, homicide, and AIDS. A recent 
analysis Of Mg,OC2XKJim^ front page coverage, reproduced in 
table 1, revealed that "The 11^ had more page-one stories about 
the dangers of flying than about any of . . . five other 
[prominent, threats to life, and on a per-death basis, it had 
orders of magnitude mere" (Barnett, 1990, . This national 
preoccupation with airline safety may provide the ultimate 
explanation for the high safety standards maintained by U.S. 
carriers and the im.,ense improvements in air safety over time. 



Barnett, Arnold, "Air Safety: End of 
the Golden Age," Chance: New 
Directions for Statistics and 
Computing . 1990, 3., 8-12. 

Barnett, Arnold, and Mary Higgins, 

"Airline Safety: The Last Decade," 

Management Science . January 1989, 
35, 1-21. 

Barnett, Arnold, and Anthony J. Lofaso, 
"After the Crash: The Passenger 
Response to the DC-10 Disaster," 
Management Science . November 1983, 
29 . 1225-1236. 

Barnett, Arnold, John Menighetti, and 

Matthew Prete, "The Public Response 
to the Sioux City DC-10 Crash," 
mimeo, 1990. 

Borenstein, Severin, "The Evolution of 
U.S. Airline Competition," mimeo, 

Borenstein, Severin, and Martin 

Zimmerman, "Market Incentives for 
Safe Commercial Airline Operation," 
American Economic Review . December 
1988, 78, 913-935. 

Chalk, Andrew, "Market Forces and 

Aircraft Safety: The Case of the 
DC-10," Economic Inquiry ^ January 
1986, 24/ 43-60. 

Chalk, Andrew, "Market Forces and 
Commercial Aircraft Safety," 
Journal of Industrial Economics . 
September 1987, 36, 61-81. 

Chance, Don M. , and Stephen P. Ferris, 
"The Effect of Aviation Disasters 
on the Air Transport Industry: A 
Financial Market Perspective," 
Journal of Transport Economics and 
Policy . May 1987, 21, 151-165. 


Evans, William N. , "Deregulation and 
Airline Safety: Evidence from 
Count Data Models," mimeo, June, 

Golbe, Devra L. , "Safety and Profits in 
the Airline Industry," Journal of 
Industrial Economics . March 1986, 
34 . 305-318. 

Kanafani, A. and Theodore E. Keeler, 
"New Entrants and Safety: Some 
Statistical Evidence on the Effects 
of Airline Deregulation." In Leon 
Moses and Ian Savage, eds. , 
Transportation Safety In an Age of 
Deregulation . Oxford: Oxford 
University Press, 1989. 

Karels, Gordon V., "Market Forces and 
Aircraft Safety: An Extension," 
Economic Inquiry . April 1989, 27 . 

Kennet, D. Mark, "Airline Deregulation 
and Aircraft Engine Maintenance: 
An Empirical Policy Analysis," 
mimeo. Fall 1990. 

McKenzie, Richard B. , and John T. 

Warner, "The Impact of Airline 
Deregulation on Highway Safety," 
mimeo, April 1988. 

Meyer, John R. , and Clinton V. Oster, 

Jr. , Deregulation and the Future of 
Intercity Passenger Travel . 
Cambridge: MIT Press, 1987. 

Mitchell, Mark L. , and Michael T. 

Maloney, "Crisis in the Cockpit? 
The Role of Market Forces in 
Promoting Air Travel Safety," 
Journal of Law and Economics . 
October 1989, 32, 329-356. 

Morrison, Steven A. , and Clifford 

Winston, "Air Safety, Deregulation, 
and Public Policy," The Brookings 
Review . Winter 1988, 6, 10-15. 

Oster, Clinton V. Jr. and C. K. Zorn, 


"Airline Deregulation: Is It Still 
Safe to Fly?." In Leon Moses and 
Ian Savage, eds., Transportation 
Safety In an Age of Deregulation . 
Oxford: Oxford University Press, 

Rose, Nancy L. "Profitability and 
Product Quality: Economic 
Determinants of Airline Safety 
Performance," Journal of Political 
Economy . October 1990, 98, 944-964. 

Rose, Nancy L. "Financial Influences 
on Airline Safety." In Leon Moses 
and Ian Savage, eds. , 
Transportation Safety In an Age of 
Deregulation . Oxford: Oxford 
University Press, 1989. 


Table 1 

Front Page Stories for Six Sources of 
Mortality Risk, 

New York Tim^c, 10/1/88 - 9/30/89 

, „ Stories 

per 1,000 

Risk Source Number of stories 
U.S. deaths 

Cancer 7 

Suicide 1 

Automobiles 4 


Homicide 35 

AIDS 35 


Commercial Jets 51 




Source: Barnett (1990), Table 4. 


Table 1 

Front Page Stories for 6 Sources of Mortality Risk, 

New York Times . 10/1/88 - 9/30/89 

Risk Source 

NuiriDer of Stories 



1.000 U.S. deaths 










Hnmi cide 






OuraiKituial Jets 



Source: Bamett (1990), Table 4. 

Figure 1 

o totrate 

A -fatrate 















25 - 

20 - 


10 - 






— 1 r 

70 75 





Aggregate Accident Rates 

Figure 2 



















20 - 

15 - 

10 - 

5 - 


Actual V. Predicted Accident Rates 

Figure 3 
















.1 - 



I r 

12 3 4 5 



^^o.^r^■K a.- ^P^^ accidents per million departures 

Distribution of Cannier Accident Rates, 1971-75 


Figure 4 







.1 - 



.05 - 




1 r 

2 3 

4 5 10 15 

total accidents per million departures 

Distribution of Carrier Accident Rates, 1981-85 


u (j t) b 



3 =1060 0075flb5fi b 

Date Due fs "^^ 



3 TOfiD 0075abSfl t>